AMD "Navi 12" Silicon Powering the Radeon Pro 5600M Rendered

[btarunr]

Out of the blue, AMD announced its Radeon Pro 5600M mobile discrete graphics solution exclusive for Apple's 16-inch MacBook Pro. It turns out that the Pro 5600M is based on an all new ASIC by AMD, codenamed "Navi 12." This is a multi-chip module, much like "Vega 20," featuring a 7 nm GPU die and two 16 Gbit (4 GB) HBM2 memory stacks sitting on an interposer. While the actual specs of the GPU die on the "Navi 14" aren't known, on the Pro 5600M, it is configured with 40 RDNA compute units amounting to 2,560 stream processors, 160 TMUs, and possibly 64 ROPs.

The engine clock of the Pro 5600M is set at up to 1035 MHz. The HBM2 memory is clocked at 1.54 Gbps, which at the 2048-bit bus width, translates to 394 GB/s of memory bandwidth. There are two big takeaways from this expensive-looking ASIC design: a significantly smaller PCB footprint compared to a "Navi 10" ASIC with its eight GDDR6 memory chips; and a significantly lower power envelope. AMD rates the typical power at just 50 W. In the render below, the new ASIC is shown next to a "Navi 14" ASIC that power RX/Pro 5500-series SKUs.



View at TechPowerUp Main Site

 

[Dante Uchiha]

All that power squeezed into 50w.

 

[Daven]

All that power squeezed into 50w.

The equivalent Nvidia solution (2070 Max-Q) is about 80W. Pretty impressive.

 

[Dante Uchiha]

The equivalent Nvidia solution (2070 Max-Q) is about 80W. Pretty impressive.

NVIDIA GeForce RTX 2070 Max-Q Specs

NVIDIA TU106, 1185 MHz, 2304 Cores, 144 TMUs, 64 ROPs, 8192 MB GDDR6, 1500 MHz, 256 bit

www.techpowerup.com

90w. Effectively almost double with the same FP32 performance.

 

[RH92]

The equivalent Nvidia solution (2070 Max-Q) is about 80W. Pretty impressive.

No not really considering the card will only clock at 1035 MHz this kind of efficiency is expected . So if i was you i would rather wait before talking about equivalence !

 

[Daven]

NVIDIA GeForce RTX 2070 Max-Q Specs

NVIDIA TU106, 1185 MHz, 2304 Cores, 144 TMUs, 64 ROPs, 8192 MB GDDR6, 1500 MHz, 256 bit

www.techpowerup.com

90w. Effectively almost double with the same FP32 performance.

Sounds about right since Nvidia is on a refined 12 nm process and AMD is on the 7 nm process (maturity level unknown). The next gen from both companies at 7 nm refined process will be very interesting.

No not really considering the card will only clock at 1035 MHz this kind of efficiency is expected . So if i was you i would rather wait before talking about equivalence !

The AMD press release talks about improved power efficiency in the title so it is fare game now. The 5700xt is just below 2000 MHz at 2560 SPs and using 225W. This 5600M Pro is just above 1000 MHz at 2560 SPs and using 50W. This is a totally unexpected result and a very large improvement in power efficiency. We can have this discussion now as we have all the information and AMD states they have improved the power efficiency today!

 

[Nioktefe]

I'm curious about the 2 small chip appart from the hbm die, the same can be found on the volta gpu, and I never found any informations about those, is it the ecc part of the hbm ?

It doesn't seem to be connected to the gpu but rather the hbm chip given the way it's shaped


Also very nice gpu, but I always wonder what kind of deal apple does with amd, there's no way they made money on vega 12, and I guess the same will be for this chip given it will surely be found only in macbooks

 

[Daven]

I'm curious about the 2 small chip appart from the hbm die, the same can be found on the volta gpu, and I never found any informations about those, is it the ecc part of the hbm ?

It doesn't seem to be connected to the gpu but rather the hbm chip given the way it's shaped


Also very nice gpu, but I always wonder what kind of deal apple does with amd, there's no way they made money on vega 12, and I guess the same will be for this chip given it will surely be found only in macbooks

The markup over the 5300M Pro is $800 so we can assume the price is over $1000. Give Apple's 35-40% margins, they may be paying AMD over $600 for the part. I think AMD is making money on it.

 

[Nioktefe]

The markup over the 5300M Pro is $800 so we can assume the price is over $1000. Give Apple's 35-40% margins, they may be paying AMD over $600 for the part. I think AMD is making money on it.

It's very pricey but still they will not sell much of them at all, so I wonder if they will manage to recoup the development cost (mask and all)

I would guess apple pay some amount upfront because all they care is having a maximum efficiency option whatever the price but maybe I'm overestimating the cost of putting the chip in production

 

[Steevo]

They used to build the small chips first like this to see what tweaks were needed for the larger size dies.

 

[Daven]

It's very pricey but still they will not sell much of them at all, so I wonder if they will manage to recoup the development cost (mask and all)

I would guess apple pay some amount upfront because all they care is having a maximum efficiency option whatever the price but maybe I'm overestimating the cost of putting the chip in production

I definitely agree that we are missing a lot of the behind the scenes info when it comes to FAB production runs, minimum orders, up front costs, etc. I would guess that it is highly unlikely that any company would make such a deal and lose money. This is the kind of thing that only happens when your company is missing from a market segment and you are trying to get your foot in the door. AMD has been providing the GPUs for Apple for years now so that market segment is not new to them.

 

[RH92]

The AMD press release talks about improved power efficiency in the title so it is fare game now. The 5700xt is just below 2000 MHz at 2560 SPs and using 225W. This 5600M Pro is just above 1000 MHz at 2560 SPs and using 50W. This is a totally unexpected result and a very large improvement in power efficiency. We can have this discussion now as we have all the information and AMD states they have improved the power efficiency today!

Oh i see you are the kind of reader that goes by what the press release Title tells you ........ so if the title tells you the moon is pink you will probably go for it i guess !

Dude press releases are press releases aka marketing , so what you are saying there means nothing . In the press release text nowhere AMD talks about improved efficiency ( hello it's the exact same architecture on the exact same node ) they talk about '' optimized efficiency for laptops '' which is very different .

All this to say that i'm not arguing with what AMD says there , i'm arguing the interpretation you make of it which is fundamentally flowed :

For starter power efficiency is exponential not linear , that means that once you go past the power efficiency sweet spot of a given architecture in order to get more clock speed the efficiency gets exponentially worse ! In other words the reason 5700XT consumes 4,5 times more power compared to the 5600M for less than double the clock speed is because visibly AMD pushed Navi10XT well beyond RDNA power efficiency sweet spot in order to hit those clock speed . So to sum this up 5600M only shows you what RDNA sweet spot is in terms of clock speed/efficiency ratio and is in no way indicative of any power efficiency improvement ( which again would be hard to explain where it came from considering we are talking about the same arch and node ) .

My second point was you made the claim 5600M equivalent is 2070MaxQ because they have around the same FP32 performance which again is fundamentally flowed . It's been demoed countless of times AMD and NVIDIA Tflops are in no way directly comparable .

Anyways i hope i made clear the misconceptions you seem to entertain and hopefully you will be able to understand them .

 

[IceShroom]

Looks like Apple buys quite good amount of AMD gpus. Even more than the pc market. That is why getting a new gpu with HBM.

 

[THANATOS]

Oh i see you are the kind of reader that goes by what the press release Title tells you ........ so if the title tells you the moon is pink you will probably go for it i guess !

Dude press releases are press releases aka marketing , so what you are saying there means nothing . In the press release text nowhere AMD talks about improved efficiency ( hello it's the exact same architecture on the exact same node ) they talk about '' optimized efficiency for laptops '' which is very different .

All this to say that i'm not arguing with what AMD says there , i'm arguing the interpretation you make of it which is fundamentally flowed :

For starter power efficiency is exponential not linear , that means that once you go past the power efficiency sweet spot of a given architecture in order to get more clock speed the efficiency gets exponentially worse ! In other words the reason 5700XT consumes 4,5 times more power compared to the 5600M for less than double the clock speed is because visibly AMD pushed Navi10XT well beyond RDNA power efficiency sweet spot in order to hit those clock speed . So to sum this up 5600M only shows you what RDNA sweet spot is in terms of clock speed/efficiency ratio and is in no way indicative of any power efficiency improvement ( which again would be hard to explain where it came from considering we are talking about the same arch and node ) .

My second point was you made the claim 5600M equivalent is 2070MaxQ because they have around the same FP32 performance which again is fundamentally flowed . It's been demoed countless of times AMD and NVIDIA Tflops are in no way directly comparable .

Anyways i hope i made clear the misconceptions you seem to entertain and hopefully you will be able to understand them .

You forgot about HBM2, which is more power efficient than GDDR6.

 

[Daven]

Oh i see you are the kind of reader that goes by what the press release Title tells you ........ so if the title tells you the moon is pink you will probably go for it i guess !

Dude press releases are press releases aka marketing , so what you are saying there means nothing . In the press release text nowhere AMD talks about improved efficiency ( hello it's the exact same architecture on the exact same node ) they talk about '' optimized efficiency for laptops '' which is very different .

All this to say that i'm not arguing with what AMD says there , i'm arguing the interpretation you make of it which is fundamentally flowed :

For starter power efficiency is exponential not linear , that means that once you go past the power efficiency sweet spot of a given architecture in order to get more clock speed the efficiency gets exponentially worse ! In other words the reason 5700XT consumes 4,5 times more power compared to the 5600M for less than double the clock speed is because visibly AMD pushed Navi10XT well beyond RDNA power efficiency sweet spot in order to hit those clock speed . So to sum this up 5600M only shows you what RDNA sweet spot is in terms of clock speed/efficiency ratio and is in no way indicative of any power efficiency improvement ( which again would be hard to explain where it came from considering we are talking about the same arch and node ) .

My second point was you made the claim 5600M equivalent is 2070MaxQ because they have around the same FP32 performance which again is fundamentally flowed . It's been demoed countless of times AMD and NVIDIA Tflops are in no way directly comparable .

Anyways i hope i made clear the misconceptions you seem to entertain and hopefully you will be able to understand them .

The only thing I need to understand is that I have been taught after many years of leaving comments on the internet is not to engage with people like you.

 

[Fluffmeister]

You forgot about HBM2, which is more power efficient than GDDR6.

Yeah the Tesla T4 is a better comparison, 16GB, 2560 CUDA cores, 70W TDP pushing 8.1 TFLOPS.

NVIDIA Tesla T4 Specs

NVIDIA TU104, 1590 MHz, 2560 Cores, 160 TMUs, 64 ROPs, 16384 MB GDDR6, 1250 MHz, 256 bit

www.techpowerup.com

 

[THANATOS]

Tesla T4 has 1590Mhz turbo, but base clock is only 585Mhz and that's only 3 TFLOPs at worst. The question is how high can It keep clocks in actual games.

 

[Fluffmeister]

Of course, but neither card is really aimed at gamers, fact is 50W TDP on this thing isn't that great, especially considering the node advantage.

 

[Vayra86]

No not really considering the card will only clock at 1035 MHz this kind of efficiency is expected . So if i was you i would rather wait before talking about equivalence !

Well... the fact remains that in direct competition the node advantage is also an economical advantage. So they are just translating die size into a huge USP. I mean 50W is really low for this kind of performance. And thát is what counts, and as long as AMD can sell that, they have an interesting product. Meanwhile we are stilll waiting on smaller node Nvidia cards to beat that.

They basically found themselves a niche.

 

[M2B]

AMD has no efficiency advantage in the mobile space.
I've seen reviews of a 100W~ 5600M losing to a 90W RTX 2060 by 10% or so.

 

[Assimilator]

AMD has no efficiency advantage in the mobile space.
I've seen reviews of a 100W~ 5600M losing to a 90W RTX 2060 by 10% or so.

Since the 5600M was literally just announced and hasn't been benchmarked yet, I'm curious as to how.

 

[IceShroom]

AMD has no efficiency advantage in the mobile space.
I've seen reviews of a 100W~ 5600M losing to a 90W RTX 2060 by 10% or so.

Arent those laptop reporting the power consumption for both the dGPU and APU, as Speedshift is always on. Also which 2060?? The desktop one or mobile one?

 

[M2B]

Since the 5600M was literally just announced and hasn't been benchmarked yet, I'm curious as to how.

 

[IceShroom]

Is that the same chip that is inside PS5?

PS5 use custom made RDNA2 chip. This Pro 5600M and 5600M is use Navi12 and Navi10 dGPU with RDNA architecture.

 

[Chrispy_]

Are we really expecting the full 2560 (40CU) configuration in something that's wearing the 5600 moniker?

Or are AMD now doing what Nvidia are doing and using "5600" as an approximate performance level, and the heavy downclock means that this will be closer to a 5600 than an 5700XT

The vanilla RX 5600 is a 32CU part clocked at 1375 so a 40CU part clocked at 1035 would offer approximately the same performance, assuming it's not bottlenecked by TDP too much.